Level Setting Presentations

The lab included several presentations which provided a variety of perspectives on how to think about the use of AI in healthcare:

• Patrick Day (BC Ministry of Health) shared an overview of the Health Sector, Information, Analysis and Reporting (HSIAR) division, and highlighted some AI projects the division is currently working on.
• Dr. Mark Schmidt (University of British Columbia) walked participants through key concepts, techniques, and current capabilities of AI, examples of how AI is currently being used today, and areas worthy of concern.
• Sarah Villeneuve (BII+E)  provided an overview of the key social and ethical considerations of AI, and provided participants with critical questions they should be asking when accessing the use and impact of an AI system.
• Shawn Gervais (Digital Technology Supercluster) spoke about how foresight methodologies could be applied to help policymakers, healthcare practitioners, and organizations think more critically about current technological developments and sectoral trends as a way to inform policy and strategy.

Group Discussions and Activities

The lab was structured to enable participants to engage with real-life case studies of AI-driven health and develop the capacity to think critically about the benefits and challenges they pose. Participants were split into groups of 4-5 people, where they were given one of five case studies. These case studies represented a range of themes across the healthcare system, including public health, public-private partnerships, mental health, health care administration, and radiology treatment. Case studies included: 

1. FINDER, a machine-learning epidemiology application that combines aggregate, anonymized Google search with location information to detect unsafe restaurants. 
2. Google Streams, an AI-driven medical assistant for clinicians to detect acute kidney injury (AKI) in patients. 
3. Tree Hole, an application that aims to identify individuals at risk of suicide by appling semantic analysis to analyze social media posts on the Chinese social media platform Weibo.
4. LKS-CHART, an advanced data analytics used to determine the optimal number of nurses to hire for the upcoming year, based on historical nurse absence data.
5. Coral Review, a software solution and a peer learning tool where radiologists can anonymously peer review medical imaging diagnoses, improving diagnostic accuracy

A facilitator guided participants through a series of activities, focused on analyzing the social, ethical, economic, and political dimensions of their AI-driven health applications. These activities prompted participants to contemplate questions such as: What individuals or groups does this application impact? Is this impact positive or negative? What assumptions guided the design of this application? What is the larger economic implication of this application? What impact may this application have on policy?

Key Themes

Several overlapping themes emerged from participant recommendations: 

1. Modernization of data regulation and privacy legislation: Current data regulation and privacy legislation may need to be revised in order to properly address concerns related to new challenges associated with privacy, surveillance, and data collection, use and sharing practices.
2. Assessment, monitoring, and maintenance of AI applications: Develop Algorithmic Impact Assessments that include Indigenous data governance perspectives, benchmarks for the tolerance of error, and require there is an established avenue for recourse if harms occur.
3. Importance of running pilot projects and following a staged approach to implementing new AI applications: Pilot projects and staged implementation processes should be informed by experts and the lived experiences of the target population. Test effectiveness and accuracy. 

Read the report for a more in-depth summary of participant recommendations, details about the day’s activities, and case study descriptions

English Version version française